Contents

This vignette explains how to performs ionomics data analysis including gene network and enrichment analysis by using the modification of R package, ionflow. The modification(ionflow_funcs) was made by Wanchang Lin () and Jacopo Iacovacci ().

0.1 Data preparation

To explore the pipeline, we’ll use the ionomics data set:

ion_data <- read.table("../test-data/iondata.tsv", header = T, sep = "\t")
dim(ion_data)
#> [1] 9999   16

Ten random lines are shown as:

sample_n(ion_data, 10)
Table 1: Samples of raw data
Knockout Batch_ID Ca Cd Co Cu Fe K Mg Mn Mo Na Ni P S Zn
YGR197C 16 135.26 0.93 0.14 1.82 10.31 3298.19 863.75 1.44 2.78 275.31 0.96 5760.13 795.04 26.44
YDL227C 6 20.35 0.67 0.11 0.89 1.50 1441.82 264.75 0.68 0.44 76.36 0.31 1693.47 61.27 12.22
YLR181C 21 45.94 1.23 0.20 2.00 9.77 3133.33 590.65 1.65 1.00 216.31 8.83 4886.48 438.14 17.99
YDL227C 89 31.76 0.88 0.14 1.46 7.90 2086.22 662.71 0.94 1.07 190.91 1.68 4141.19 429.41 11.75
YJL192C 28 45.49 1.55 0.17 1.75 9.22 2606.80 681.01 1.55 1.30 217.81 1.62 4814.28 596.28 16.48
YLR133W 21 36.90 0.90 0.19 1.95 10.25 2942.86 600.78 1.30 1.10 181.62 1.40 4492.62 489.04 15.79
YKL106W 23 36.73 0.99 0.20 1.71 7.60 2301.65 647.38 0.97 0.77 157.70 1.45 4320.44 442.54 15.19
YDL227C 91 22.14 0.76 0.11 0.85 5.60 2706.70 672.84 0.88 1.30 203.78 0.56 4277.77 467.11 13.96
YJL144W 29 47.58 1.17 0.16 1.82 9.66 2854.36 769.41 1.25 1.23 250.44 1.64 5149.42 555.72 16.90
YLR165C 80 13.23 1.02 0.15 3.10 7.73 2563.36 804.38 0.96 1.55 237.82 1.30 5634.82 448.21 16.36

The first few columns are meta information such as gene ORF and batch id. The rest is the ionomics data.

0.2 Data pre-process

The raw data set should be pre-processed. The pre-processing function PreProcessing performs:

The raw data are at first log trainsformed and then followed by the batch correction. The user can chose not to perform batch correction, otherwise the user can use either median or median plus std method. If there is quality control for the batch correction, the user can use it and indicates in the argument of control_lines. Also this function gives user option how to use these control line (control_use): If control_use is control, these control lines (data rows) are used for the batch correction factor; if control.out, lines except control lines are used.

This data set has a control line: YDL227C mutant. The code segment below is to identify it:

max(with(ion_data, table(Knockout)))
#> [1] 1617
which.max(with(ion_data, table(Knockout)))
#> YDL227C 
#>     209

The next stage is outlier detection. Here only univarite methods are implemented, including mad, IQR, and log.FC.dist. And like batch correction, user can skip this procedure by setting method_outliers = none in the function argument. There is a threshold to control the number of outliers. The larger the threshold (thres_outl) the more outlier removal.

Standarisation provides three methods: std, mad or custom. If the method is cumstom, user must use specific std values such as:

std <- read.table("../test-data/user_std.tsv", header = T, sep = "\t")
std
#>    Ion     sd
#> 1   Ca 0.1508
#> 2   Cd 0.0573
#> 3   Co 0.0580
#> 4   Cu 0.0735
#> 5   Fe 0.1639
#> 6    K 0.0940
#> 7   Mg 0.0597
#> 8   Mn 0.0771
#> 9   Mo 0.1142
#> 10  Na 0.1075
#> 11  Ni 0.0784
#> 12   P 0.0597
#> 13   S 0.0801
#> 14  Zn 0.0671

The pre-process procedure returns not only processed ionomics data but also a symbolic data set. This data set is based on the inomics data and is determined by a threshold(thres_symb):

The core part of network and enrivhment analysis, clustering, is based on the symbolic data.

Let’s run the pre-process procedure:

pre <- PreProcessing(data = ion_data,
                     var_id = 1, batch_id = 2, data_id = 3,
                     method_norm = "median",
                     control_lines = "YDL227C",
                     control_use = "control",
                     method_outliers = "IQR",
                     thres_outl = 3,
                     stand_method = "std",
                     stdev = NULL,
                     thres_symb = 3)

names(pre)
#> [1] "stats.raw_data"    "stats.outliers"    "stats.batch_data" 
#> [4] "data.long"         "data.gene.logFC"   "data.gene.zscores"
#> [7] "data.gene.symb"    "plot.dot"          "plot.hist"

The results includes summaries of raw data and processed data. The latter is:

pre$stats.batch_data %>% 
  kable(caption = 'Processed data summary', digits = 2, booktabs = T) %>%
  kable_styling(full_width = F, font_size = 10)
Table 2: Processed data summary
Ion Min. 1st Qu. Median Mean 3rd Qu. Max. Variance
Ca -4.45 -0.28 -0.13 -0.12 0.02 2.35 0.11
Cd -1.70 0.03 0.10 0.11 0.17 0.93 0.03
Co -2.80 0.02 0.09 0.06 0.15 1.60 0.05
Cu -0.66 -0.10 -0.03 -0.01 0.04 5.28 0.04
Fe -7.48 -0.17 -0.06 -0.02 0.07 6.88 0.14
K -2.21 -0.17 -0.01 -0.08 0.09 1.83 0.08
Mg -1.84 -0.06 0.01 -0.01 0.07 1.69 0.03
Mn -4.11 -0.24 -0.08 -0.13 0.01 1.78 0.06
Mo -2.03 -0.26 -0.08 -0.08 0.09 4.44 0.13
Na -7.41 -0.53 -0.22 -0.33 -0.04 1.25 0.24
Ni -2.40 -0.01 0.09 0.12 0.21 7.90 0.12
P -1.18 -0.06 0.00 -0.01 0.06 1.45 0.02
S -2.38 -0.03 0.05 0.06 0.16 2.38 0.04
Zn -0.46 -0.08 -0.03 -0.01 0.03 4.60 0.02

The pre-processed data and symbolic data are like like:

pre$data.gene.zscores %>% head() %>%
  kable(caption = 'Processed data', digits = 2, booktabs = T) %>% 
  kable_styling(full_width = F, font_size = 10,
                latex_options = c("striped", "scale_down"))
Table 3: Processed data
Line Ca Cd Co Cu Fe K Mg Mn Mo Na Ni P S Zn
YAL004W -1.16 0.75 1.19 -0.47 0.04 0.61 0.51 -0.84 -0.08 -1.84 1.71 0.52 0.33 -0.09
YAL005C -1.67 0.84 0.55 0.58 -2.79 0.59 0.31 -1.16 -1.42 -0.12 1.48 0.73 0.13 -0.13
YAL007C -2.12 0.64 0.23 -0.53 -0.24 0.79 -0.09 -0.14 1.22 -0.92 0.00 0.09 -0.29 -0.65
YAL008W -2.34 1.13 0.21 -0.73 -2.16 0.52 -0.02 -0.87 0.93 -0.58 0.02 -0.09 -0.73 -0.47
YAL009W -1.18 0.66 0.55 -1.11 -3.91 0.22 0.09 -0.18 1.50 -0.84 -0.09 0.14 0.01 -0.36
YAL010C -1.28 1.43 2.27 0.46 1.53 -2.75 0.04 -0.74 -9.71 -4.30 2.42 -0.98 -0.05 -0.01 

pre$data.gene.symb %>% head() %>%
  kable(caption = 'Symbolic data', booktabs = T) %>%
  kable_styling(full_width = F, font_size = 10)
Table 3: Symbolic data
Line Ca Cd Co Cu Fe K Mg Mn Mo Na Ni P S Zn
YAL004W 0 0 0 0 0 0 0 0 0 0 0 0 0 0
YAL005C 0 0 0 0 0 0 0 0 0 0 0 0 0 0
YAL007C 0 0 0 0 0 0 0 0 0 0 0 0 0 0
YAL008W 0 0 0 0 0 0 0 0 0 0 0 0 0 0
YAL009W 0 0 0 0 -1 0 0 0 0 0 0 0 0 0
YAL010C 0 0 0 0 0 0 0 0 -1 -1 0 0 0 0

The symbolic data are calulated from the processed data with control of thres_symb (here is 3). You can obtain a new symbol data set by re-assigning a new threshold to the function symbol_data:

data_symb <- symbol_data(pre$data.gene.zscores, thres_symb = 2)
data_symb %>% head() %>%
  kable(caption = 'Symbolic data with threshold of 2', booktabs = T) %>%
  kable_styling(full_width = F, font_size = 10)
Table 4: Symbolic data with threshold of 2
Line Ca Cd Co Cu Fe K Mg Mn Mo Na Ni P S Zn
YAL004W 0 0 0 0 0 0 0 0 0 0 0 0 0 0
YAL005C 0 0 0 0 -1 0 0 0 0 0 0 0 0 0
YAL007C -1 0 0 0 0 0 0 0 0 0 0 0 0 0
YAL008W -1 0 0 0 -1 0 0 0 0 0 0 0 0 0
YAL009W 0 0 0 0 -1 0 0 0 0 0 0 0 0 0
YAL010C 0 0 1 0 0 -1 0 0 -1 -1 1 0 0 0

The pre-processed data distribution is:

pre$plot.hist
Ionomcs data distribution plot

Figure 1: Ionomcs data distribution plot

0.3 Data filtering

There are a lot of ways to filter genes. Here we filter genes based on symbolic data: remove genes with all velues are zero.

data <- pre$data.gene.zscores
data_symb <- pre$data.gene.symb
idx <- rowSums(abs(data_symb[, -1])) > 0
dat <- data[idx, ]
dat_symb <- data_symb[idx, ]
dim(dat)
#> [1] 549  15

0.4 Data clustering

The hierarchical cluster analysis is the key part of gene network and gene enrichment analysis. The methodology is as follow:

One example is:

clust <- gene_clus(dat_symb[, -1], min_clust_size = 10)
names(clust)
#> [1] "clus"    "idx"     "tab"     "tab_sub"

The cluster centres are:

clust$tab_sub
#>   cluster nGenes
#> 1       4    149
#> 2      11     72
#> 3       7     36
#> 4       1     27
#> 5      18     15
#> 6       5     12
#> 7       3     11
#> 8       8     11

It indicates that clusters and their number of genes (larger than min_cluster_size).

0.5 Gene network

The gene network uses both the ionomics and symboloc data. The similarity measures on the ionomics data are filtered by the similarity threshold located between 0 and 1, and cluster centres of symbolic data. The filter values are then used for network analysis.

The similarity measure method is one of pearson, spearman, kendall, cosine, mahal_cosine or hybrid_mahal_cosine. For the last two methods, see publication: Extraction and Integration of Genetic Networks from Short-Profile Omic Data Sets for details.

For example, we use the Pearson correlation as similarity measure for netwok analysis:

net <- GeneNetwork(data = dat,
                   data_symb = dat_symb,
                   min_clust_size = 10,
                   thres_corr = 0.75,
                   method_corr = "pearson")

The network with nodes coloured by the symbolic data clustering is:

net$plot.pnet1
Netwok analysis based on Pearson correlation: symbolic clustering

Figure 2: Netwok analysis based on Pearson correlation: symbolic clustering

The same network, but nodes are colured by the netwok community detection:

net$plot.pnet2
Netwok analysis based on Pearson correlation: community detction

Figure 3: Netwok analysis based on Pearson correlation: community detction

The network analysis also returns a network impact and betweeness plot:

net$plot.impact_betweenness
Netwok analysis based on Pearson correlation: impact and betweeness

Figure 4: Netwok analysis based on Pearson correlation: impact and betweeness

For the comparision purpose, we use different similarity methods. Here we choose Cosine:

net_1 <- GeneNetwork(data = dat,
                     data_symb = dat_symb,
                     min_clust_size = 10,
                     thres_corr = 0.75,
                     method_corr = "cosine")
net_1$plot.pnet1
Netwok analysis based on Cosine

Figure 5: Netwok analysis based on Cosine

net_1$plot.pnet2
Netwok analysis based on Cosine

Figure 6: Netwok analysis based on Cosine

Use Hybrid Mahalanobis Cosine:

net_2 <- GeneNetwork(data = dat,
                     data_symb = dat_symb,
                     min_clust_size = 10,
                     thres_corr = 0.75,
                     method_corr = "mahal_cosine")
net_2$plot.pnet1
Netwok analysis based on Mahalanobis Cosine

Figure 7: Netwok analysis based on Mahalanobis Cosine

net_2$plot.pnet2
Netwok analysis based on Mahalanobis Cosine

Figure 8: Netwok analysis based on Mahalanobis Cosine

Again, we use Hybrid Mahalanobis Cosine:

net_3 <- GeneNetwork(data = dat,
                     data_symb = dat_symb,
                     min_clust_size = 10,
                     thres_corr = 0.75,
                     method_corr = "hybrid_mahal_cosine")
net_3$plot.pnet1
Netwok analysis based on Hybrid Mahalanobis Cosine

Figure 9: Netwok analysis based on Hybrid Mahalanobis Cosine

net_3$plot.pnet2
Netwok analysis based on Hybrid Mahalanobis Cosine

Figure 10: Netwok analysis based on Hybrid Mahalanobis Cosine

0.6 Enrichment analysis

The KEGG enrichment analysis:

kegg <- kegg_enrich(data = dat_symb, min_clust_size = 10, pval = 0.05,
                    annot_pkg =  "org.Sc.sgd.db")

#' kegg
kegg %>% 
  kable(caption = 'KEGG enrichmenat analysis', digits = 3, booktabs = T) %>%
  kable_styling(full_width = F, font_size = 10,
                latex_options = c("striped", "scale_down"))
Table 5: KEGG enrichmenat analysis
Cluster KEGGID Pvalue Count Size Term
Cluster 7 (36 genes) 03010 0.029 9 16 Ribosome
Cluster 7 (36 genes) 00330 0.031 3 3 Arginine and proline metabolism
Cluster 18 (15 genes) 00290 0.009 2 2 Valine, leucine and isoleucine biosynthesis
Cluster 18 (15 genes) 00520 0.009 2 2 Amino sugar and nucleotide sugar metabolism
Cluster 18 (15 genes) 00260 0.012 3 6 Glycine, serine and threonine metabolism
Cluster 18 (15 genes) 00010 0.024 2 3 Glycolysis / Gluconeogenesis
Cluster 18 (15 genes) 01110 0.037 5 22 Biosynthesis of secondary metabolites
Cluster 3 (11 genes) 00400 0.009 2 2 Phenylalanine, tyrosine and tryptophan biosynthesis
Cluster 8 (11 genes) 01100 0.006 6 55 Metabolic pathways
Cluster 8 (11 genes) 00564 0.027 2 6 Glycerophospholipid metabolism

Note that there can be none results for KRGG enrichment analysis. Change arguments such as thres_clus as appropriate.

The GO Terms enrichment analysis:

go <- go_enrich(data = dat_symb, min_clust_size = 10, pval = 0.05,
                ont = "BP", annot_pkg =  "org.Sc.sgd.db")
#' go
go %>% head() %>%
  kable(caption = 'GO Terms enrichmenat analysis', digits = 3, booktabs = T) %>%
  kable_styling(full_width = F, font_size = 10,
                latex_options = c("striped", "scale_down"))
Table 6: GO Terms enrichmenat analysis
Cluster ID Description Pvalue Count CountUniverse Ontology
Cluster 4 (149 genes) GO:0051336 regulation of hydrolase activity 0.0018 4 12 BP
Cluster 4 (149 genes) GO:0043085 positive regulation of catalytic activity 0.0044 4 15 BP
Cluster 4 (149 genes) GO:0035303 regulation of dephosphorylation 0.0068 2 3 BP
Cluster 4 (149 genes) GO:0046889 positive regulation of lipid biosynthetic process 0.0068 2 3 BP
Cluster 4 (149 genes) GO:1903727 positive regulation of phospholipid metabolic process 0.0068 2 3 BP
Cluster 4 (149 genes) GO:0044764 multi-organism cellular process 0.0074 3 9 BP

0.7 Exploratory analysis

Some analysis are performed in terms of ions, i.e. feature, including PCA and correlation.

expl <- ExploratoryAnalysis(data = dat)
Exploratory analysis plots with respect to ionome

Figure 11: Exploratory analysis plots with respect to ionome

Exploratory analysis plots with respect to ionome

Figure 12: Exploratory analysis plots with respect to ionome

Exploratory analysis plots with respect to ionome

Figure 13: Exploratory analysis plots with respect to ionome

expl$plot.PCA_Individual
Exploratory analysis plots with respect to ionome

Figure 14: Exploratory analysis plots with respect to ionome

expl$plot.correlation_network
Exploratory analysis plots with respect to ionome

Figure 15: Exploratory analysis plots with respect to ionome